Grant Proposal (RFP-9): Ghost AI / Ghost Agent - Anonymous Web3 Notary

I am applying for a $5k Developer Grant.

Please review my application and discuss! See DFINITY’s directions for becoming a registered reviewer here. They will be collected by DFINITY. When two weeks passes, DFINITY will release them and they will appear as a new section on this post.

Please review my application and discuss! If you would like to submit an official review, then please use the links below to register, see the rubric, and submit a review.

I’m looking forward to everyone’s input!

Reviewer Registration | Rubric for Evaluating | Submit a Review

MY APPLICATION:

REVIEWS:

Proto here: ghost.grondon.com
Project Name: Ghost AI / Ghost Agent
Applicant: Gabriel R
Forum Handle: @Gabrielrondon
Grant Amount: $5,000
Category: Decentralized AI (DeAI)
Project GitHub: Ghost AI GitHub
Prototype Frontend: Ghost AI Prototype
Documentation: Ghost AI Docs

Project Description

Ghost AI is a decentralized, privacy-preserving notary system that enables users to prove they have taken specific actions without revealing their identity. Using Zero-Knowledge Proofs (ZKPs) and AI-driven context awareness, Ghost AI creates verifiable attestations that allow users to prove their participation in Web3 governance, transactions, and off-chain events without exposing sensitive data.

Ghost AI is an open-source project with a long-term vision of enabling privacy-focused attestations for Web3 users. Our initial MVP focuses on two core functionalities:

MVP Task List

1. Generate Anonymous Proof of Action

  • Users submit an action to Ghost AI (e.g., transaction, vote, event).
  • Ghost AI verifies the action using blockchain or off-chain data.
  • A Zero-Knowledge Proof (ZKP) is generated to confirm the action without exposing the user.
  • Users receive a verifiable proof link that third parties can check.

Example Tasks:

  • Generate anonymous proof of DAO vote participation.
  • Prove that a user staked tokens without revealing their wallet.
  • Verify event attendance without exposing personal identity.

2. Provide Human-Readable Proof Summary (AI-Powered)

  • AI analyzes and explains the proof in human-readable language.
  • Instead of raw blockchain data, Ghost AI returns an easy-to-understand summary.

Example Tasks:

  • Summarize proof of participation in a DAO proposal.
  • Explain anonymous staking proof in simple terms.
  • Provide a human-readable verification summary for third-party validation.

Milestones & Roadmap

Milestone 1: MVP Development ($2,500) – 4 weeks

  • Deploy a canister of our notary system on ICP.
  • Implement anonymous reference generation for attestation requests.
  • Develop a simple frontend for submitting proof requests.
  • Integrate basic ZK proof attestations for verified actions.
  • Test execution with sample use cases.

Milestone 2: Privacy & Security Enhancements ($2,500) – 6 weeks

  • Strengthen the ZK proof system for improved anonymity and verifiability.
  • Add multi-chain execution capabilities via ICP’s Chain Fusion.
  • Implement additional AI-driven verification capabilities.
  • Release an open-source implementation and documentation.

Impact on the Internet Computer Ecosystem

  • Expands ICP’s decentralized AI capabilities with privacy-preserving attestations.
  • Introduces a unique application of Zero-Knowledge proofs for anonymous Web3 verification.
  • Enhances DAO governance, DeFi compliance, and Web3 education credentialing.

Future Vision

Ghost AI’s long-term goal is to enable selective proof disclosure, secure storage, and third-party verification functionalities. These features will be developed in future iterations as the project grows.

We invite the community to contribute and provide feedback to shape the evolution of Ghost AI. Looking forward to discussions on how to make this a core part of the ICP ecosystem!

PS: The project is open-source

1 Like

Extension (updated in our Gitbook)

The problem we aim to tackle: Ghost AI Agent addresses the fundamental tension between privacy and verification in Web3. Currently, users must choose between revealing their entire identity/activity or sacrificing verifiability. This creates significant barriers:

  • When participating in DAO governance, users expose their entire wallet history
  • Users can’t prove specific attributes without revealing unrelated personal data
  • Cross-chain verification remains cumbersome and exposes multiple identities
  • Few solutions offer configurable levels of privacy disclosure

Ghost Agent value to the Internet Computer ecosystem by enabling privacy-preserving interactions with both ICP and external chains, creating accessible ZK-proof tools that don’t require specialized expertise. And as this is an open-source project, the vision is that this can be the seed of something bigger.

I’m happy to elaborate on any technical aspects or implementation details!

PS: My recent experience with ZK, was explaining for regulators in Brazil the advantages of it. I’ve created a very simple demo (with simple circuits on Circom): zkp.gabrielrondon.com
Recently I am more involved with the Noir & Semaphore community, researching and building stuff there. (zkVote: GitHub - gabrielrondon/zkvote)

PS2: On the Internet Computer-side, More than one year ago I produced an audio-series (icp) on Spotify. Due to budget cuts, the voices were made by AI, however all the scriptwriting process were made by humans. This project connected me with the ICP Portuguese Hub.

PS3: Regarding AI, in this grant we are using it to automate a few things. An agent that by reading inputs can translate (with privacy) what is going on. Also it is the beginning of the automated tasks (by creating playlists). (GitHub - gabrielrondon/ai_agent: for educational purposes)

Technical Implementation Details

I’d like to share some specifics about how Ghost Agent leverages AI within its privacy-preserving architecture:

Ghost Agent uses a dual-layer approach:

  1. Core privacy layer: Rust canister on ICP for ZK proof generation/verification and secure task execution
  2. Task automation layer: AI-powered orchestration for managing privacy-preserving workflows

The AI component specifically:

  • Interprets user task intentions and configures the appropriate ZK circuits
  • Manages the execution flow across different blockchains via Chain Fusion
  • Generates human-readable attestations from complex blockchain data while preserving privacy

Our implementation focuses on practical automation rather than advanced natural language capabilities - think of it as an “agent” in the sense of a privacy-preserving automation layer.

The code I’ve shared demonstrates the core reference generation, task management, and execution components. With this grant, I’ll complete the ZK proof integration and multi-chain execution features.

Happy to elaborate on any technical aspects you’d like to know more about!